Massive time-series datasets from sensor networks, scientific monitoring, and online services pose a challenge for interactive visual exploration, as the number of samples can exceed screen pixels by several orders of magnitude. Naïve subsampling harms interpretability and event detection, and existing pixel-aware methods usually operate at a single resolution, failing to exploit the multi-resolution nature of zoom and pan workflows. We propose Adaptive Multi-Resolution Stratified Sampling (AMRSS), which combines stratified min–max sampling with a hierarchical multi-resolution index and an adaptive query-time selection algorithm. AMRSS precomputes a summary tree whose nodes store min–max envelopes and a local variability score; for a given viewport and pixel budget, a top-down traversal selects variable-resolution segments that concentrate samples in visually complex regions while avoiding oversampling in flat regions, yielding a sample set bounded by the screen width and renderable as a polyline with predictable complexity. We formalize pixel-aware sampling under viewport constraints, present the AMRSS design and its time–space complexity, and outline an evaluation protocol against uniform subsampling, fixed-resolution min–max, and triangle-based methods on seismic, physiological, and synthetic datasets, providing a practical basis for interactive dashboards and a framework extensible to formal error bounds, streaming updates, and perceptual studies.
Massive time-series datasets from sensor networks, scientific monitoring, and online services pose a challenge for interactive visual exploration, as the number of samples can exceed screen pixels by several orders of magnitude. Naïve subsampling harms interpretability and event detection, and existing pixel-aware methods usually operate at a single resolution, failing to exploit the multi-resolution nature of zoom and pan workflows. We propose Adaptive Multi-Resolution Stratified Sampling (AMRSS), which combines stratified min–max sampling with a hierarchical multi-resolution index and an adaptive query-time selection algorithm. AMRSS precomputes a summary tree whose nodes store min–max envelopes and a local variability score; for a given viewport and pixel budget, a top-down traversal selects variable-resolution segments that concentrate samples in visually complex regions while avoiding oversampling in flat regions, yielding a sample set bounded by the screen width and renderable as a polyline with predictable complexity. We formalize pixel-aware sampling under viewport constraints, present the AMRSS design and its time–space complexity, and outline an evaluation protocol against uniform subsampling, fixed-resolution min–max, and triangle-based methods on seismic, physiological, and synthetic datasets, providing a practical basis for interactive dashboards and a framework extensible to formal error bounds, streaming updates, and perceptual studies.
